From 81b6b09983ab21cb07ad443b8d9b08debe81c599 Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Thu, 3 Jan 2013 14:30:04 +0100
Subject: [PATCH] Make buffer-to-buffer enqueue_copy safe by default if no
byte_count specified.
---
doc/source/runtime.rst | 68 +----------------------------
pyopencl/__init__.py | 78 ++++++++++++++++++++++++++++++++--
src/wrapper/wrap_cl.hpp | 11 +++--
src/wrapper/wrap_cl_part_1.cpp | 2 +-
4 files changed, 85 insertions(+), 74 deletions(-)
diff --git a/doc/source/runtime.rst b/doc/source/runtime.rst
index 547ce018..ccfa1172 100644
--- a/doc/source/runtime.rst
+++ b/doc/source/runtime.rst
@@ -526,73 +526,7 @@ Images
Transfers
^^^^^^^^^
-.. function:: enqueue_copy(queue, dest, src, **kwargs)
-
- Copy from :class:`Image`, :class:`Buffer` or the host to
- :class:`Image`, :class:`Buffer` or the host. (Note: host-to-host
- copies are unsupported.)
-
- The following keyword arguments are available:
-
- :arg wait_for: (optional, default empty)
- :arg is_blocking: Wait for completion. Defaults to *True*.
- (Available on any copy involving host memory)
-
- :return: A :class:`NannyEvent` if the transfer involved a
- host-side buffer, otherwise an :class:`Event`.
-
- :class:`Buffer` ↔ host transfers:
-
- :arg device_offset: offset in bytes (optional)
-
- :class:`Buffer` ↔ :class:`Buffer` transfers:
-
- :arg byte_count: (optional)
- :arg src_offset: (optional)
- :arg dest_offset: (optional)
-
- Rectangular :class:`Buffer` ↔ host transfers (CL 1.1 and newer):
-
- :arg buffer_origin: :class:`tuple` of :class:`int` of length
- three or shorter. (mandatory)
- :arg host_origin: :class:`tuple` of :class:`int` of length
- three or shorter. (mandatory)
- :arg region: :class:`tuple` of :class:`int` of length
- three or shorter. (mandatory)
- :arg buffer_pitches: :class:`tuple` of :class:`int` of length
- two or shorter. (optional, "tightly-packed" if unspecified)
- :arg host_pitches: :class:`tuple` of :class:`int` of length
- two or shorter. (optional, "tightly-packed" if unspecified)
-
- :class:`Image` ↔ host transfers:
-
- :arg origin: :class:`tuple` of :class:`int` of length
- three or shorter. (mandatory)
- :arg region: :class:`tuple` of :class:`int` of length
- three or shorter. (mandatory)
- :arg pitches: :class:`tuple` of :class:`int` of length
- two or shorter. (optional)
-
- :class:`Buffer` ↔ :class:`Image` transfers:
-
- :arg offset: offset in buffer (mandatory)
- :arg origin: :class:`tuple` of :class:`int` of length
- three or shorter. (mandatory)
- :arg region: :class:`tuple` of :class:`int` of length
- three or shorter. (mandatory)
-
- :class:`Image` ↔ :class:`Image` transfers:
-
- :arg src_origin: :class:`tuple` of :class:`int` of length
- three or shorter. (mandatory)
- :arg dest_origin: :class:`tuple` of :class:`int` of length
- three or shorter. (mandatory)
- :arg region: :class:`tuple` of :class:`int` of length
- three or shorter. (mandatory)
-
- |std-enqueue-blurb|
-
- .. versionadded:: 2011.1
+.. autofunction:: enqueue_copy(queue, dest, src, **kwargs)
Mapping Memory into Host Address Space
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
diff --git a/pyopencl/__init__.py b/pyopencl/__init__.py
index 59628fca..f7a4a535 100644
--- a/pyopencl/__init__.py
+++ b/pyopencl/__init__.py
@@ -1,3 +1,5 @@
+# -*- coding: utf-8 -*-
+
__copyright__ = "Copyright (C) 2009 Andreas Kloeckner"
__license__ = """
@@ -420,7 +422,7 @@ def _add_functionality():
pass
what = e.what + "\n\n" + (75*"="+"\n").join(
- "Build on %s:\n\n%s" % (dev, log)
+ "Build on %s:\n\n%s" % (dev, log)
for dev, log in self._get_build_logs())
code = e.code
routine = e.routine
@@ -580,7 +582,7 @@ def _add_functionality():
def image_format_repr(self):
return "ImageFormat(%s, %s)" % (
- channel_order.to_string(self.channel_order,
+ channel_order.to_string(self.channel_order,
"<unknown channel order 0x%x>"),
channel_type.to_string(self.channel_data_type,
"<unknown channel data type 0x%x>"))
@@ -598,7 +600,7 @@ def _add_functionality():
except AttributeError:
return str(val)
else:
- result = "%s failed: %s" % (val.routine(),
+ result = "%s failed: %s" % (val.routine(),
status_code.to_string(val.code(), "<unknown error %d>")
.lower().replace("_", " "))
if val.what():
@@ -809,6 +811,76 @@ if _cl.get_cl_header_version() >= (1,1):
enqueue_copy_buffer_rect = _mark_copy_deprecated(_cl._enqueue_copy_buffer_rect)
def enqueue_copy(queue, dest, src, **kwargs):
+ """Copy from :class:`Image`, :class:`Buffer` or the host to
+ :class:`Image`, :class:`Buffer` or the host. (Note: host-to-host
+ copies are unsupported.)
+
+ The following keyword arguments are available:
+
+ :arg wait_for: (optional, default empty)
+ :arg is_blocking: Wait for completion. Defaults to *True*.
+ (Available on any copy involving host memory)
+
+ :return: A :class:`NannyEvent` if the transfer involved a
+ host-side buffer, otherwise an :class:`Event`.
+
+ :class:`Buffer` ↔ host transfers:
+
+ :arg device_offset: offset in bytes (optional)
+
+ :class:`Buffer` ↔ :class:`Buffer` transfers:
+
+ :arg byte_count: (optional) If not specified, defaults to the
+ size of the source in versions 2012.x and earlier,
+ and to the minimum of the size of the source and target
+ from 2013.1 on.
+ :arg src_offset: (optional)
+ :arg dest_offset: (optional)
+
+ Rectangular :class:`Buffer` ↔ host transfers (CL 1.1 and newer):
+
+ :arg buffer_origin: :class:`tuple` of :class:`int` of length
+ three or shorter. (mandatory)
+ :arg host_origin: :class:`tuple` of :class:`int` of length
+ three or shorter. (mandatory)
+ :arg region: :class:`tuple` of :class:`int` of length
+ three or shorter. (mandatory)
+ :arg buffer_pitches: :class:`tuple` of :class:`int` of length
+ two or shorter. (optional, "tightly-packed" if unspecified)
+ :arg host_pitches: :class:`tuple` of :class:`int` of length
+ two or shorter. (optional, "tightly-packed" if unspecified)
+
+ :class:`Image` ↔ host transfers:
+
+ :arg origin: :class:`tuple` of :class:`int` of length
+ three or shorter. (mandatory)
+ :arg region: :class:`tuple` of :class:`int` of length
+ three or shorter. (mandatory)
+ :arg pitches: :class:`tuple` of :class:`int` of length
+ two or shorter. (optional)
+
+ :class:`Buffer` ↔ :class:`Image` transfers:
+
+ :arg offset: offset in buffer (mandatory)
+ :arg origin: :class:`tuple` of :class:`int` of length
+ three or shorter. (mandatory)
+ :arg region: :class:`tuple` of :class:`int` of length
+ three or shorter. (mandatory)
+
+ :class:`Image` ↔ :class:`Image` transfers:
+
+ :arg src_origin: :class:`tuple` of :class:`int` of length
+ three or shorter. (mandatory)
+ :arg dest_origin: :class:`tuple` of :class:`int` of length
+ three or shorter. (mandatory)
+ :arg region: :class:`tuple` of :class:`int` of length
+ three or shorter. (mandatory)
+
+ |std-enqueue-blurb|
+
+ .. versionadded:: 2011.1
+ """
+
if isinstance(dest, MemoryObjectHolder):
if dest.type == mem_object_type.BUFFER:
if isinstance(src, MemoryObjectHolder):
diff --git a/src/wrapper/wrap_cl.hpp b/src/wrapper/wrap_cl.hpp
index 6c64ac97..c71e33f6 100644
--- a/src/wrapper/wrap_cl.hpp
+++ b/src/wrapper/wrap_cl.hpp
@@ -1845,17 +1845,22 @@ namespace pyopencl
command_queue &cq,
memory_object_holder &src,
memory_object_holder &dst,
- size_t byte_count,
+ ptrdiff_t byte_count,
size_t src_offset,
size_t dst_offset,
py::object py_wait_for)
{
PYOPENCL_PARSE_WAIT_FOR;
- if (byte_count == 0)
+ if (byte_count < 0)
{
+ size_t byte_count_src = 0;
+ size_t byte_count_dst = 0;
PYOPENCL_CALL_GUARDED(clGetMemObjectInfo,
- (src.data(), CL_MEM_SIZE, sizeof(byte_count), &byte_count, 0));
+ (src.data(), CL_MEM_SIZE, sizeof(byte_count), &byte_count_src, 0));
+ PYOPENCL_CALL_GUARDED(clGetMemObjectInfo,
+ (src.data(), CL_MEM_SIZE, sizeof(byte_count), &byte_count_dst, 0));
+ byte_count = std::min(byte_count_src, byte_count_dst);
}
cl_event evt;
diff --git a/src/wrapper/wrap_cl_part_1.cpp b/src/wrapper/wrap_cl_part_1.cpp
index d441718f..b9f55a06 100644
--- a/src/wrapper/wrap_cl_part_1.cpp
+++ b/src/wrapper/wrap_cl_part_1.cpp
@@ -228,7 +228,7 @@ void pyopencl_expose_part_1()
py::return_value_policy<py::manage_new_object>());
py::def("_enqueue_copy_buffer", enqueue_copy_buffer,
(py::args("queue", "src", "dst"),
- py::arg("byte_count")=0,
+ py::arg("byte_count")=-1,
py::arg("src_offset")=0,
py::arg("dst_offset")=0,
py::arg("wait_for")=py::object()
--
GitLab